AI Lab

28/07/2026

The models you choose make a big difference. Do your homework before making any long-term commitments.

24/07/2026

Don’t just blindly automate your company’s work with whatever LLM you’re used to. This graph is a perfect illustration of why it’s important to pick the right model for every task. The cost difference over the years will be huge.

22/07/2026

Everything you can do in the Publisher UI is now available through a versioned REST API.

Articles, AI drafting, deep research, publishing and WordPress sync, social posts, approval workflows, schedules, AI timeslot proposals, media, the knowledge base, team and roles, analytics, platform connections. All scriptable.

API keys are organisation-scoped, act as a specific team member, and intersect that member's permissions. So UI capability and API capability stay in lockstep automatically.

Agencies and teams can feed Publisher from their own tools, build custom dashboards, and automate publishing end to end. Standard conventions: status polling for long-running jobs, per-key rate limits, additive-only contract.

Publisher has changed from a tool for writing and scheduling into a data-driven content engine. The platform now tracks real-world performance across your accounts, feeds those insights directly into an

16/07/2026

Mixture-of-experts is the cheapest path to high intelligence per dollar. Only the relevant expert sub-network fires per token. The model behaves like a smaller specialist on every call.

Many companies are blowing through their budgets on hungry AI tools that burn cash with wordy, inefficient answers. To keep costs under control, businesses must stop looking at flat pricing

15/07/2026

The Commission's high-risk AI classification guidelines are open for stakeholder feedback. If your AI is on the boundary, this is where you push back.

Details:

On 19 May 2026 the European Commission published draft guidelines on how to classify high-risk AI systems under Article 6 of the AI Act. Consultation is open until 23 June

14/07/2026

12-16 extra months on high-risk AI compliance. Three things to do with the runway:
1. Recheck classification
2. Build, not defer
3. Verify SMC threshold

The AI Omnibus provisional agreement has now cleared Coreper and Parliament committee approval, with formal adoption still pending. For AI builders, the working deal delays high-risk AI obligations, gives SMEs

13/07/2026

Small software teams are going to look bigger.

Not because agents remove engineers, but because each engineer can supervise more useful work: research, code, tests, docs, migrations.

The advantage shifts to fast judgment.

Anthropic's recursive self-improvement essay is most useful as an engineering signal, not a doomsday forecast. Its internal data suggests code writing is becoming less scarce, while specification, review, testing, and

09/07/2026

Three levers cut AI cost-per-outcome in production. Pick the right size model. Measure tokens at the application layer. Route cheap-by-default with premium escalation. Stop reading the pricing page.

Many companies are blowing through their budgets on hungry AI tools that burn cash with wordy, inefficient answers. To keep costs under control, businesses must stop looking at flat pricing

08/07/2026

Article drafting in Publisher is now run end to end by an autonomous agent, not a fixed pipeline.

You give it your source material. Notes, links, files. It studies everything, decides when to search the web for current facts, reviews your tone of voice, pulls from your knowledge base, and writes the draft. It runs in the background with a live progress view, so you can close the tab and come back to it.

A fixed pipeline does the same steps in the same order every time. An agent adapts. Researches when the facts are not on hand. Skips work the topic does not need. And keeps a memory of its decisions, so editing the draft becomes a conversation rather than a restart.

Publisher has changed from a tool for writing and scheduling into a data-driven content engine. The platform now tracks real-world performance across your accounts, feeds those insights directly into an

08/07/2026

In Play Text your text never leaves your device. Not because we promise to look away, but because there is no server for it to go to. Voices generated on-device, no telemetry, no accounts.

That matters if you read confidential documents to yourself.

We built Play Text, a free and private text-to-speech app for Mac and iOS, in just two evenings. It turns articles, PDFs, and web pages into audio with live, word-by-word